Inspiration
Recovering from an injury is hard to do alone. I got injured as a beginner while working out, and one of my teammates also injured his leg. When we tried existing fitness apps, they mostly offered generic exercise listsbut not the part we needed most: real-time guidance on how to move safely and correctly while recovering.
We’re building Align for beginners, people recovering from injuries, and older adultsanyone who wants safer movement with clear feedback.
So we built Align as a rehab companion: something that can watch your movement, understand your recovery context, and give clear coaching cues you can act on immediately.
What it does
Align is a rehab-focused movement coach for beginners, people recovering from injuries, and older adults:
- Real-time movement feedback: analyzes your movement and generates short, actionable form/gait cues.
- Personalized coaching: uses your profile context (goals, constraints, injury notes) to tailor guidance.
- Spoken cues (hands-free): coaching is read out loud so you can stay focused on the exercise, not the screen.
- Multiple voice avatars: create several “coach voices,” including a voice cloned from your own recording, and choose which one speaks your coaching.
- Doctor workflow (Swift-only, no backend):
- Patient exports a share file to send to a clinician.
- Clinician imports it in “Doctor Mode,” edits the rehab plan + adds instructions, then exports an update file.
- Patient imports the update file back into Align.
- Clinician notes are high priority: doctor instructions are injected into the AI context as “HIGH PRIORITY constraints” so coaching follows them first.
How we built it
iOS app (SwiftUI) for the end-to-end rehab experience:
- onboarding + profile editing
- a live coaching flow that generates movement summaries and plays spoken cues
- on iOS, in-app voice recording to capture a sample for cloning
- offline Doctor Mode for import/edit/export of rehab plans + clinician notes
Generative AI + voice stack
- Google Gemini API generates rehab-style coaching guidance from the movement summary + profile context.
- ElevenLabs API generates spoken coaching via Text-to-Speech, and supports instant voice cloning (voice avatars via
voice_id).
Challenges we ran into
- Latency vs. usefulness: real-time rehab cues need to feel immediate, so we kept requests lean while preserving enough context to be helpful.
- Safety and tone: injury recovery is sensitive. We focused on practical cues without medical claims.
- Audio UX reliability: recording → uploading → cloning → playing feedback required careful state management and permissions (microphone).
- Clinician alignment without a backend: we built a file-based workflow so doctors can review and edit safely offline.
Accomplishments that we're proud of
- A working rehab-first coaching loop that produces actionable cues instead of generic plans.
- Multiple voice avatars (including voice cloning) so coaching feels personal and motivating.
- A Swift-only clinician workflow (export/import) that makes it easy for doctors to add instructions without setting up infrastructure.
- Clinician instructions treated as high-priority constraints in the AI context.
What we learned
- Recovery isn’t just “do these exercises” it’s technique, consistency, and confidence.
- The best coaching is short, specific, and timed to what the user is doing.
- Personal spoken feedback can make rehab feel less isolating and easier to stick with.
What's next for Align
- Rehab programs & progression: plans that adapt over time, not just single-session cues.
- Better measurement: clearer rehab metrics, trends, and improvement tracking.
- Clinician UX polish: richer plan templates (sets/reps/time), better import/export UX, and clearer safety guardrails.
- Privacy controls: stronger controls for what gets shared in the patient package.
Log in or sign up for Devpost to join the conversation.